Você está na página 1de 5

Life assessment of turbine components based on deterministic

and probabilistic procedures


Albert Bagaviev
*
, Artur Ulbrich
Siemens Power Generation, Mulheim/Ruhr, Germany
Abstract
Long life endurance and high availability are very important factors in the operation of large steam turbines. The operational safety is
ensured by regular inspections in combination with mechanical calculations. In this context, risk-based life assessment approaches, which
balance out safety and economic requirements, have become more and more important in the last years (e.g. the developing a unied
approach to make risk-based decisions within inspection and maintenance in the frame of the project Risk based Inspection and Maintenance
Procedure for European Industry (RIMAP)).
The article shortly outlines the theoretical background of life assessment with the help of probabilistic fracture mechanics and
demonstrates its application for typically heavy-loaded components of large stream turbines.
q 2004 Elsevier Ltd. All rights reserved.
Keywords: Risk-based component life assessment; Risk based inspection and maintenance procedure for European industry; Finite element method;
Probabilistic fracture mechanics; Steam turbines
1. Introduction
Risk-based life assessment approaches, which balance
out safety and economic requirements, have become more
and more important in the last years. One of the activities in
this context, is the development of a unied approach to
make risk-based decisions within inspection and mainten-
ance, the European Project Risk based Inspection and
Maintenance Procedure for European Industry (RIMAP).
One of the main benets of the project is a creation of a
technical framework for a European standardisation in the
area of risk-based maintenance and inspection planning. In
this content one of the work packages is dedicated to the
development of detailed quantied risk assessment methods.
The article demonstrates an application of the risk
assessment methods in the power industry in the frame of
the probabilistic fracture mechanics.
As in many other industrial elds, it is becoming more
and more important in the operation of large power plants to
balance out safety and economical requirements. Naturally,
every operator strives for long inspection intervals, which in
turn require a very accurate calculation of damage process
and the corresponding life endurance. Here it is essential to
avoid unnecessary high safety factors while at the same time
the operational safety has to be guaranteed. Hence, it is
essential to identify unnecessary high safety factors in
components dimensions, allowable crack sizes or com-
ponents inspection intervals.
Modern calculation methods like Finite Element Method
(FEM) help to further approximate to the actual component
limits compared to classical methods. At the same time, this
demands to estimate possible risks and impacts of failure as
accurate are possible, which in turn increased the appli-
cation of the probabilistic methods in the engineering eld.
Generally, the components are designed by deterministic
methods rst; based on that, failure probabilities are
calculated and assessed (i.e. nally accepted or rejected)
with respect to the corresponding failure impacts.
The article in hand illustrates this area with the example
of the assessment of intervals in which the shrunk-un disks
of steam turbine rotors are inspected. Section 2 introduces
some material failure limits that are typical in mechanical
turbine engineering. Section 3 gives an overview of
methods to calculate the failure probability. In Section 4
0308-0161/$ - see front matter q 2004 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ijpvp.2004.07.003
International Journal of Pressure Vessels and Piping 81 (2004) 855859
www.elsevier.com/locate/ijpvp
* Corresponding author.
E-mail address: albert.bagaviev@seimens.com (A. Bagaviev).
the applied probability distribution of material parameters,
loads, etc. are given, while Section 5 shows the obtained
results for the above-mentioned disks.
2. Failure assessment methods
The turbine component integrity evaluation uses the
well-known failure assessment diagram (FAD) method [2].
This approach enables integrity analysis of components
with defects at the same time due to brittle fracture and
plastic collapse. The FAD locus divides the area of the co-
ordinate frame (S
r
, K
r
) into safe and unsafe regions, See.
Fig. 1.
Once the brittle fracture parameter K
r
Z
K
I
K
IC
and the
plastic collapse parameter S
r
Z
s
s
Y
for the case under
investigation are calculated, the relative position of the
evaluation point (S
r
, K
r
) is considered regarding the FAD
locus.
Taking into account the main failure mechanism, three
limit state (failure) functions are used:
Steam turbine rotor and shrunk-on disks (nuclear units)
design excludes in wide range of operational speed the
plastic collapse occurrence. Therefore, the reduced
failure assessment is predominantly based on the linear
elastic fracture mechanics (LEFM) with small plasticity
effect correction at the crack tip (so called small scale
yielding), see Fig. 1 (FAD
SSY
line). Thus, the resulting
limit state function has the following form
JK
IC
; s
Y
; a; x; s; geometry
ZK
IC
KK
I
s
Y
; a; x; s; geometry
Low-pressure rotors and modern shrunk-on disks are
made of materials with high ductility. This leads to
the necessity of application the limit state function based
on the J-Integral calculation in the frame of the crack
driving force (CDF) concept, see Fig. 2
JJ
IC
; s
Y
; a; x; s; geometry
ZJ
IC
KJs
Y
; a; x; s; geometry
Stationary steam turbine components like cylinder
casings, valve casings and piping have a common
featurethey have a nite geometry (wall thickness)
regarding the system crackremaining ligament, so
that the concept of plastic collapse can be applied to
characterize the remaining through-the-wall plastica-
tion. Limit state function is based on the failure
assessment diagram, see Fig. 1 (FAD line)
JK
IC
; s
Y
; a; x; s ZK
r
K
pS
r

8 ln sec
p
2
S
r
_ _ _ _
_
3. Probabilistic failure assessment
Denote by x
1
; .; x
N
an N-dimensional random vector,
characterising in the frame of the assumed model the
uncertainties concerning the geometry, material properties,
applied loads and existing/assumed defects which can be
modelled as random variables.
The probability of failure is dened as
P
f
Z
_
Jx
1
;.;x
N
%0
f x
1
; .; x
N
dx
1
; .; dx
N
where J(x
1
,.,x
N
) is the limit state function (failure
surface), f(x
1
,.,x
N
) is the joint probability density function
of (x
1
,.,x
N
).
There are some commonly used approximating methods
available for performing the integration of the above
multidimensional integral in probabilistic uncertainty
analysis.
Fig. 1. FAD diagram.
Fig. 2. Failure assessment based on the J-integral method.
A. Bagaviev, A. Ulbrich / International Journal of Pressure Vessels and Piping 81 (2004) 855859 856
4. Mean value estimation method (MVM)
This method typically involves developing the Taylor
series expansion of limit state function J(x
1
,.,x
N
) about
the mean values x
i
of random uncorrelated variables iZ
1,.,N. The rst and second moments of resulting
approximation are then used to calculate the failure
probability [1]
Jx
1
; .; x
N
zJ x
1
; .; x
N
C

N
iZ1
vJ x
1
; .; x
N

vx
i
x
i
K x
i

C
1
2!

N
iZ1

N
jZ1
v
2
J x
1
; .; x
N

vx
i
vx
j
!x
i
K x
i
x
j
K x
N

The expectation of this expression or the rst moment about


the mean is then

Jx
1
;.;x
N
ZEJx
1
;.;x
N
zJ x
1
;.; x
N

C
1
2!

N
iZ1

N
jZ1
v
2
J x
1
;.; x
N

vx
i
vx
j
x
i
K x
i
x
j
K x
N

The same procedure applied to the expression for the second


moment or variance yields
VJx
1
;.;x
N
z

N
iZ1
Vx
i

vJ x
1
;.; x
N

vx
i
_ _
2
C2

N
iZ1

N
jZ1
Covx
i
;x
j

vJ x
1
;.; x
N

vx
i
_ _
!
vJ x
1
;.; x
N

vx
j
_ _
Under assumption of normality of the limit state function
J(x
1
,.,x
N
) an evaluation of the failure probability can be
made
PrfJx
1
;.;x
N
!0g
ZPr
Jx
1
;.;x
N
KEJx
1
;.;x
N

VJx
1
;.;x
N

p !
0KEJx
1
;.;x
N

VJx
1
;.;x
N

p
_ _
P
f
zF K
EJx
1
;.;x
N

VJx
1
;.;x
N

p
_ _
where F() is the standard normal cumulative density
function.
5. First-order reliability methods (FORM)
This method can be considered as an extension of the
mean value estimation method. The rst-order approxi-
mation of the limit state function (Taylor series expansion
about the design point) is considered at the closest surface
point (most probable point (MPP) or design point) to the
origin of the standard Gaussian space.
First, the original random variables (x
1
,.x
N
) are
transformed into independent standard Gaussian variables
(u
1
,.u
N
). The original limit state function is then mapped
into the new limit state function (J
U
(u
1
,.,u
N
)Z0.
Second, the point on the surface (in standard Gaussian
space) with the shortest distance to the origin is determined.
This point is referred to as design point and the distance
from the design point to the origin is called reliability index
b. The probability of failure P
f
is thus approximated by
PrfJ
U
u
1
; .; u
N
!0g ZFKb
6. Monte Carlo simulation method
To evaluate the multidimensional integral
P
f
Z
_
Jx
1
;.;x
N
%0
f x
1
; .; x
N
dx
1
; .; dx
N
;
n realisations of the random vector of model variables x
1
1
; .; x
1
N
; .; x
n
1
; .; x
n
N
are generated. Corresponding
output samples J
1
x
1
1
; .; x
1
N
; .; J
n
x
n
1
; .; x
n
N
are
counted as follows
Pz
1
n

n
iZ1
UJx; .; x%0
where
UJx
1
; .; x
N
%0 Z
1 if Jx
1
; .; x
N
%0;
0 otherwise
_
The approximation approaches the exact value of failure
probability when the number n of trials approaches innity.
An enhancement of the crude Monte Carlo method is the
Monte Carlo simulation with importance sampling. Using
FORM the design or most probable point can be obtained
and the samples are taken around the important MPP with
importance-sampling density.
7. Random variables
In this section, the used probability distributions are
given. Further for the application example it is assumed that
the variables are uncorrelated.
7.1. Fracture toughness
The fracture toughness follows a three parameter
Weibull distribution with the density
A. Bagaviev, A. Ulbrich / International Journal of Pressure Vessels and Piping 81 (2004) 855859 857
f K
Ic
Z
a
h
K
Ic
Kg
h
_ _
aK1
exp K
K
Ic
Kg
h
_ _ _ _
with position, shape and scale parameters.
7.2. Yield strength
For the yield strength values a distribution with a
lognormal density is assumed
f s
Y
Z
1
s
Y
s

2p
p exp K
lns
Y
Km
2
2s
2
_ _
7.3. Crack growth rate
A model known as Westinghouse-rate with mean value
of
ln _ a ZK1:733 K
4056:7
T C273:3
C0:004032s
Y
and standard deviation of 0.587 is found to describe
appropriately a crack growth due to the stress corrosion in
nuclear power plants [7] as a lognormal variable.
7.4. Crack initiation probability
Based on the eld data about the number of indications
found during inspections of number of different disk types
and different locations, the crack initiation probability q
i
can
be estimated as a binomial distributed variable.
7.5. Crack shape probability
The crack shape description based on the crack depth to
crack length ratio is assumed to be distributed normally with
mean of 0.4 and with the standard deviation of 0.1.
7.6. Applied Load
For the description of the applied load randomness a
normal distribution with a coefcient of variation of 0.1 is
taken. The mean value is calculated based on the results of
the two-dimensional Finite Element (FE) modelling [3], see.
Fig. 3.
8. Application: probability of disc rupture due to stress
corrosion
Disk-type rotors are used in large power plant turbines
with large exhaust steam volumes, i.e. generally in nuclear
power plants. Fig. 3 shows a low-pressure turbine with an
exhaust area of 20 m
2
per ow, corresponding to the more
recent Siemens/KWU models.
The spindle shaft accommodates eight shrunk-on disks.
Although stress corrosion cracking is conservatively
assumed to take place in older-type turbines right from the
start, it should be mentioned that it is possible to reduce the
danger of SCC initiation in newer turbines almost
arbitrarily. Experimental and eld investigations [46]
have shown that, even under severe steam conditions, no
stress corrosion cracks are initiated when the ratio of
(tensile) surface stress and yield strength remains below 0.5.
This requirement was fullled throughout the newer
Siemens PG disk-type rotors.
For the service example considered here the possibility
of SCC should be conservatively taken into account.
Moreover, it should be assumed that the disks are made of
older type material (with lower fracture toughness and high
yield strength values), so that the method based on LEFM
will be used.
Thus, failure occurs when the maximum value of stress
intensity factor K
I
exceeds the fracture toughness K
Ic
, i.e.
JK
Ic
; s
Y
; x; k; a
0
; _ a; s
Z
K
Ic
sapplied load; geometry

p
Q
_
Ygeometry
_ _
2
Kat
1
; t
2
%0
for an operational interval between t
1
and t
2
.
For shrunk-on disks the most signicant crack growth
mechanism is stress corrosion cracking (SCC). Low cycle
fatigue due to the starts-ups and shut-downs processes is
Fig. 3. Low pressure steam turbine and tangential stress distribution in a
shrunk-on disk.
A. Bagaviev, A. Ulbrich / International Journal of Pressure Vessels and Piping 81 (2004) 855859 858
insignicant because of the low number of starts due to the
base-load operation character of nuclear units. For stress
corrosion cracking the following evolution equation is
assumed
a Za
0
C
_
t
2
t
1
_ as
Y
; Tdt;
whereby as the cases found in the literature on this topic
shows, the assumed stress corrosion cracks grow typically in
geometrically self-similar manner, i.e. the crack depth/-
crack length ratio can be ignored.
Fig. 4 shows the evolution of the disc rupture probability
due to the stress corrosion crack growth during operation
estimated with the mean value method (MVM line), rst
order reliability method (FORM line) and Monte Carlo
simulation method with importance sampling (MCSIS line).
The MCSIS is assumed to be a method yielding exact
results.
9. Conclusion
A methodology was outlined for the evaluation of failure
probability of shrunk-on disk rupture due to stress corrosion
cracking. The basis is an elastic analysis by FEM to
compute the stress distribution in the turbine part, the linear
elastic fracture mechanics approach within small scale
yielding to formulate the failure surface and standard
computational methods of structural reliability theory.
The failure probabilities were computed using MVM,
FORM and MCS with importance sampling.
The results show a very good agreement between the
Monte Carlo simulation method and FORM. The FORM is
much less time-consuming compared to the MCS method.
Thus, the FORM provides a fast and effective method for
computing failure probabilities of steam turbine parts and
can be used instead of currently used time-consuming MC
simulation for calculation of the turbine inspection
intervals.
References
[1] Hahn GJ, Shapiro SS. Statistical models in engineering. New York:
Wiley; 1967.
[2] Milne I, Ainsworth RA, Dowling AR, Stewart AT. Assessment of the
integrity of structures containing defects. Int J Pressure Vessel Piping
1988;32(3).
[3] Harder J, Bagaviev A, Luft H. Increase of availability of steam turbines
by means of non-linear fracture mechanics. in: ASME, PVP-Vol. 431
2001. Book No. G01180-2001.
[4] David W, Roettger G, Schleichoff K. Disk-type LP turbine rotor
experience. International Joint Power Generation Conference, Kansas
City, Missouri 1993.
[5] Termuellen H, Schleithoff K, Neumann K. Advanced Disk-type LP
Turbine Rotors EPRI Workshop on Stress Corrosion Cracking in Steam
Turbines, Charlotte, North Carolina, October 1011, 1990.
[6] Oeynhausen H, Roettger G, Ewald J, Schleithoff K, Termuellen H.
Reliable Disk-type Rotors for Nuclear Power Plants, American Power
Conference, Chicago, Illinois, April 2729, 1987.
[7] Hodge JM, Mogford IL. UK experience of stress corrosion cracking in
steam turbine disc. Proc Inst Mech Engr 1979;193(11).
Fig. 4. Effect of operational hours on probability of disk rupture according to different probabilistic methods.
A. Bagaviev, A. Ulbrich / International Journal of Pressure Vessels and Piping 81 (2004) 855859 859

Você também pode gostar